UnifyBench

RankingEffort benchmarks

GPT-5.3 Codex effort benchmarks

Compare measured Low, Medium, High and other settings. Every number keeps its source.

GPT-5.3 Codex

12 measured results · 1 reported effort labels · 12 benchmark/harness combinations. Model evidence and sources →

Capability score evidence

GPT-5.3 Codex · XHigh: 23.0 overall comparison score. Meets the overall evidence requirements.

Which sources carry the weight?
  • Artificial Analysis · 81.5% direct comparison weight · independent
  • SWE-rebench · 18.5% direct comparison weight · independent

These shares describe the comparisons entering the fit, not fractions of the final score.

How sensitive is this score?

Remove one evaluator or family and refit the graph against the same overall panel. Values use the matching dataset audited 2026-09-12. These are sensitivity checks, not confidence intervals; ranking eligibility may also change.

Removed evidenceRecomputed scoreChange
Artificial Analysis4.8-18.2
ARC Prize22.6-0.5
Datacurve20.6-2.4
LiveCodeBench22.9-0.1
Scale24.6+1.6
Terminal-Bench23.0+0.0
Cognition24.0+1.0
LiveBench22.8-0.2
SWE-rebench33.0+9.9
Epoch AI23.3+0.3
Vals AI22.5-0.5
Provider reports · Anthropic23.0+0.0
Provider reports · Moonshot AI23.0+0.0
Provider reports · Meta23.0-0.0
Provider reports · Mistral23.0+0.0
All provider reports23.0-0.0
Every benchmark family
Removed evidenceRecomputed scoreChange
swe-rebench33.0+9.9
deepswe20.6-2.4
tau-bench24.2+1.2
mmmu-pro24.2+1.1
frontiercode24.0+1.0
omniscience22.1-0.9
aa-lcr22.1-0.9
briefcase23.8+0.7
epoch-game-puzzles22.3-0.7
simpleqa23.6+0.6
gpqa22.5-0.6
arc-agi22.5-0.5
critpt22.6-0.5
mmlu-pro23.4+0.4
scicode23.3+0.3
livebench-instructions23.3+0.3
multi-swe-bench22.8-0.2
hle23.3+0.2
gdp-pdf23.2+0.2
swe-atlas-qna22.8-0.2
apex-agents23.2+0.2
harvey22.9-0.2
livebench-language23.2+0.1
frontiermath23.2+0.1
arc-agi-322.9-0.1
livebench-reasoning22.9-0.1
livebench-coding22.9-0.1
vibe-code22.9-0.1
gdpval23.1+0.1
ifbench22.9-0.1
gmmlu23.1+0.1
vals-finance-agent23.0-0.1
analyst-agent23.1+0.1
vals-legal-research23.0-0.1
terminal-bench23.0-0.1
vals-code-migration23.0-0.1
swe-atlas-test-writing23.0-0.0
enigma-eval23.1+0.0
livecodebench23.0-0.0
math50023.1+0.0
livebench-data23.1+0.0
enterprise-ops23.1+0.0
vals-excel-modeling23.0-0.0
swe-atlas-refactoring23.0-0.0
livebench-math23.0+0.0
automationbench23.0+0.0
aime23.0+0.0
swe-bench-pro23.0+0.0
itbench23.0-0.0
What evidence is missing from the fit?

Across all collected settings for this model: 12 observations, 11 contributing, 0 matched without graph weight and 1 excluded. Counts do not establish rank eligibility.

  • 1 · Benchmark family or source protocol has not been reviewed
Inspect every observation and exclusion →

Capability profile

One configuration, seven capabilities

Comparison score · 0–100
GPT-5.3 Codex · XHigh capability radarComparison scores from 0 to 100. Unknown capabilities have no point; lines stop at gaps. Hollow points are preliminary. Exact values and support labels follow the chart.AgenticHard reasoningCodingHuman prefKnowledgeMultimodalLong context50100GPT-5.3 Codex · XHigh · Agentic: 20.9 · SupportedGPT-5.3 Codex · XHigh · Hard reasoning: 45.9 · SupportedGPT-5.3 Codex · XHigh · Coding: 22.9 · SupportedGPT-5.3 Codex · XHigh · Knowledge: 43.0 · SupportedGPT-5.3 Codex · XHigh · Multimodal: 33.7 · PreliminaryGPT-5.3 Codex · XHigh · Long context: 67.5 · Preliminary

Gaps are unknown, not zero. Hollow points are preliminary.

  • Agentic

    A20.9Supported

  • Hard reasoning

    A45.9Supported

  • Coding

    A22.9Supported

  • Human pref

    Unknown

  • Knowledge

    A43.0Supported

  • Multimodal

    A33.7Preliminary

  • Long context

    A67.5Preliminary

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Compare this configuration →Explore the model evidence profile →

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Benchmark / sourceXHigh
AA-Briefcase EloArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

AA-Briefcase is an agentic knowledge work benchmark developed by Artificial Analysis. AA-Briefcase Elo is a combined metric that aggregates rubric pass rate, analytical quality Elo and presentation Elo · Higher is better · Evaluation results measured independently by Artificial Analysis

Unit: elo. Source collected 2026-09-12.

865.74 Elo
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

83.33%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

CritPtArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

16.86%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

91.52%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

42.49%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

IFBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

75.37%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

MMMU-ProArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

78.5%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

AA-Omniscience IndexArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis

Unit: index. Source collected 2026-09-12.

10.87
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

AA-Omniscience AccuracyArtificial Analysis · 6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader
Benchmark details

AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

52.88%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

𝜏²-Bench TelecomArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

85.96%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

Terminal-Bench HardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

53.03%
Reported settings

GPT-5.3 Codex (xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex

SWE-rebench · 2026-02 · toolsSWE-rebench · SWE-rebench standardized ReAct tools
Benchmark details

Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.

Unit: percent. Source collected 2026-09-12.

58.6%
Reported settings

SWE-rebench standardized ReAct tools; swe-rebench-2026-02-tools; gpt-5.3-codex-xhigh; checkpoint=gpt-5.3-codex

What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 7,424 results across 240 models · Collected 2026-09-12